#define DEBUG_SUBSYSTEM S_RPC
#ifdef __KERNEL__
-#include <linux/config.h>
-#include <linux/module.h>
-#include <linux/kmod.h>
+# include <linux/config.h>
+# include <linux/module.h>
+# include <linux/kmod.h>
#else
-#include <liblustre.h>
+# include <liblustre.h>
#endif
#include <linux/obd_support.h>
#include <linux/lustre_import.h>
#include <linux/lustre_export.h>
#include <linux/obd.h>
+#include <linux/obd_ost.h>
#include <linux/obd_class.h>
#include <linux/obd_lov.h> /* for IOC_LOV_SET_OSC_ACTIVE */
#include "ptlrpc_internal.h"
-int ptlrpc_reconnect_import(struct obd_import *imp,
- struct ptlrpc_request **reqptr)
-{
- struct obd_device *obd = imp->imp_obd;
- int flags, rc, size[] = {sizeof(imp->imp_target_uuid),
- sizeof(obd->obd_uuid),
- sizeof(imp->imp_dlm_handle)};
- char *tmp[] = {imp->imp_target_uuid.uuid,
- obd->obd_uuid.uuid,
- (char *)&imp->imp_dlm_handle};
- struct ptlrpc_connection *conn = imp->imp_connection;
- struct ptlrpc_request *req;
- struct lustre_handle old_hdl;
-
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_generation++;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- CERROR("reconnect handle "LPX64"\n",
- imp->imp_dlm_handle.cookie);
-
- req = ptlrpc_prep_req(imp, imp->imp_connect_op, 3, size, tmp);
- if (!req)
- RETURN(-ENOMEM);
- req->rq_level = LUSTRE_CONN_NEW;
- req->rq_replen = lustre_msg_size(0, NULL);
- rc = ptlrpc_queue_wait(req);
- if (rc) {
- CERROR("cannot connect to %s@%s: rc = %d\n",
- imp->imp_target_uuid.uuid, conn->c_remote_uuid.uuid, rc);
- GOTO(out_disc, rc);
- }
-
- if (lustre_msg_get_op_flags(req->rq_repmsg) & MSG_CONNECT_RECONNECT) {
- memset(&old_hdl, 0, sizeof(old_hdl));
- if (!memcmp(&old_hdl, &req->rq_repmsg->handle,
- sizeof (old_hdl))) {
- CERROR("%s@%s didn't like our handle "LPX64
- ", failed\n", imp->imp_target_uuid.uuid,
- conn->c_remote_uuid.uuid,
- imp->imp_dlm_handle.cookie);
- GOTO(out_disc, rc = -ENOTCONN);
- }
-
- if (memcmp(&imp->imp_remote_handle, &req->rq_repmsg->handle,
- sizeof(imp->imp_remote_handle))) {
- CERROR("%s@%s changed handle from "LPX64" to "LPX64
- "; copying, but this may foreshadow disaster\n",
- imp->imp_target_uuid.uuid,
- conn->c_remote_uuid.uuid,
- imp->imp_remote_handle.cookie,
- req->rq_repmsg->handle.cookie);
- imp->imp_remote_handle = req->rq_repmsg->handle;
- GOTO(out_disc, rc = 0);
- }
-
- CERROR("reconnected to %s@%s after partition\n",
- imp->imp_target_uuid.uuid, conn->c_remote_uuid.uuid);
- GOTO(out_disc, rc = 0);
- }
-
- old_hdl = imp->imp_remote_handle;
- imp->imp_remote_handle = req->rq_repmsg->handle;
- CERROR("reconnected to %s@%s ("LPX64", was "LPX64")!\n",
- imp->imp_target_uuid.uuid, conn->c_remote_uuid.uuid,
- imp->imp_remote_handle.cookie, old_hdl.cookie);
- GOTO(out_disc, rc = 0);
-
- out_disc:
- *reqptr = req;
- return rc;
-}
+static int ptlrpc_recover_import_no_retry(struct obd_import *, char *);
void ptlrpc_run_recovery_over_upcall(struct obd_device *obd)
{
char *argv[4];
char *envp[3];
int rc;
-
ENTRY;
+
argv[0] = obd_lustre_upcall;
argv[1] = "RECOVERY_OVER";
- argv[2] = obd->obd_uuid.uuid;
+ argv[2] = (char *)obd->obd_uuid.uuid;
argv[3] = NULL;
-
+
envp[0] = "HOME=/";
envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
envp[2] = NULL;
rc = USERMODEHELPER(argv[0], argv, envp);
if (rc < 0) {
CERROR("Error invoking recovery upcall %s %s %s: %d; check "
- "/proc/sys/lustre/upcall\n",
+ "/proc/sys/lustre/upcall\n",
argv[0], argv[1], argv[2], rc);
-
+
} else {
- CERROR("Invoked upcall %s %s %s",
- argv[0], argv[1], argv[2]);
+ CWARN("Invoked upcall %s %s %s\n",
+ argv[0], argv[1], argv[2]);
}
}
void ptlrpc_run_failed_import_upcall(struct obd_import* imp)
{
- char *argv[6];
+#ifdef __KERNEL__
+ unsigned long flags;
+ char *argv[7];
char *envp[3];
int rc;
-
ENTRY;
+
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ if (imp->imp_state == LUSTRE_IMP_CLOSED) {
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+ EXIT;
+ return;
+ }
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+
argv[0] = obd_lustre_upcall;
argv[1] = "FAILED_IMPORT";
- argv[2] = imp->imp_target_uuid.uuid;
- argv[3] = imp->imp_obd->obd_uuid.uuid;
- argv[4] = imp->imp_connection->c_remote_uuid.uuid;
- argv[5] = NULL;
+ argv[2] = (char *)imp->imp_target_uuid.uuid;
+ argv[3] = imp->imp_obd->obd_name;
+ argv[4] = (char *)imp->imp_connection->c_remote_uuid.uuid;
+ argv[5] = (char *)imp->imp_obd->obd_uuid.uuid;
+ argv[6] = NULL;
envp[0] = "HOME=/";
envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
rc = USERMODEHELPER(argv[0], argv, envp);
if (rc < 0) {
- CERROR("Error invoking recovery upcall %s %s %s %s %s: %d; check "
- "/proc/sys/lustre/lustre_upcall\n",
+ CERROR("Error invoking recovery upcall %s %s %s %s %s: %d; "
+ "check /proc/sys/lustre/upcall\n",
argv[0], argv[1], argv[2], argv[3], argv[4],rc);
-
+
} else {
- CERROR("Invoked upcall %s %s %s %s %s\n",
- argv[0], argv[1], argv[2], argv[3], argv[4]);
+ CWARN("Invoked upcall %s %s %s %s %s\n",
+ argv[0], argv[1], argv[2], argv[3], argv[4]);
+ }
+#else
+ if (imp->imp_state == LUSTRE_IMP_CLOSED) {
+ EXIT;
+ return;
}
+ ptlrpc_recover_import(imp, NULL);
+#endif
}
-int ptlrpc_replay(struct obd_import *imp)
+/* This might block waiting for the upcall to start, so it should
+ * not be called from a thread that shouldn't block. (Like ptlrpcd) */
+void ptlrpc_initiate_recovery(struct obd_import *imp)
+{
+ ENTRY;
+
+ LASSERT (obd_lustre_upcall != NULL);
+
+ if (strcmp(obd_lustre_upcall, "DEFAULT") == 0) {
+ CDEBUG(D_HA, "%s: starting recovery without upcall\n",
+ imp->imp_target_uuid.uuid);
+ ptlrpc_connect_import(imp, NULL);
+ }
+ else if (strcmp(obd_lustre_upcall, "NONE") == 0) {
+ CDEBUG(D_HA, "%s: recovery disabled\n",
+ imp->imp_target_uuid.uuid);
+ }
+ else {
+ CDEBUG(D_HA, "%s: calling upcall to start recovery\n",
+ imp->imp_target_uuid.uuid);
+ ptlrpc_run_failed_import_upcall(imp);
+ }
+
+ EXIT;
+}
+
+int ptlrpc_replay_next(struct obd_import *imp, int *inflight)
{
int rc = 0;
struct list_head *tmp, *pos;
- struct ptlrpc_request *req;
+ struct ptlrpc_request *req = NULL;
unsigned long flags;
- __u64 committed = imp->imp_peer_committed_transno;
+ __u64 last_transno;
ENTRY;
+ *inflight = 0;
+
/* It might have committed some after we last spoke, so make sure we
* get rid of them now.
*/
spin_lock_irqsave(&imp->imp_lock, flags);
ptlrpc_free_committed(imp);
+ last_transno = imp->imp_last_replay_transno;
spin_unlock_irqrestore(&imp->imp_lock, flags);
- CDEBUG(D_HA, "import %p from %s has committed "LPD64"\n",
- imp, imp->imp_target_uuid.uuid, committed);
-
- list_for_each(tmp, &imp->imp_replay_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
- DEBUG_REQ(D_HA, req, "RETAINED: ");
- }
+ CDEBUG(D_HA, "import %p from %s committed "LPU64" last "LPU64"\n",
+ imp, imp->imp_target_uuid.uuid, imp->imp_peer_committed_transno,
+ last_transno);
/* Do I need to hold a lock across this iteration? We shouldn't be
* racing with any additions to the list, because we're in recovery
* than the one we're replaying (it can't be committed until it's
* replayed, and we're doing that here). l_f_e_safe protects against
* problems with the current request being committed, in the unlikely
- * event of that race. So, in conclusion, I think that it's safe to
+ * event of that race. So, in conclusion, I think that it's safe to
* perform this list-walk without the imp_lock held.
*
* But, the {mdc,osc}_replay_open callbacks both iterate
* just a little race...
*/
list_for_each_safe(tmp, pos, &imp->imp_replay_list) {
- req = list_entry(tmp, struct ptlrpc_request, rq_list);
+ req = list_entry(tmp, struct ptlrpc_request, rq_replay_list);
+
+ /* If need to resend, stop on the matching one first. It's
+ possible though it's already been committed, so in that case
+ we'll just continue with replay */
+ if (imp->imp_resend_replay &&
+ req->rq_transno == last_transno) {
+ lustre_msg_add_flags(req->rq_reqmsg, MSG_RESENT);
+ break;
+ }
- DEBUG_REQ(D_HA, req, "REPLAY:");
+ if (req->rq_transno > last_transno) {
+ imp->imp_last_replay_transno = req->rq_transno;
+ break;
+ }
+
+ req = NULL;
+ }
+ imp->imp_resend_replay = 0;
+
+ if (req != NULL) {
rc = ptlrpc_replay_req(req);
-
if (rc) {
- CERROR("recovery replay error %d for req "LPD64"\n",
- rc, req->rq_xid);
+ CERROR("recovery replay error %d for req "
+ LPD64"\n", rc, req->rq_xid);
RETURN(rc);
}
+ imp->imp_reqs_replayed++;
+ *inflight = 1;
}
-
- RETURN(0);
+ RETURN(rc);
}
int ptlrpc_resend(struct obd_import *imp)
* list, so we don't need to hold the lock during this iteration and
* resend process.
*/
+ /* Well... what if lctl recover is called twice at the same time?
+ */
spin_lock_irqsave(&imp->imp_lock, flags);
- LASSERT(imp->imp_level < LUSTRE_CONN_FULL);
+ if (imp->imp_state != LUSTRE_IMP_RECOVER) {
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+ RETURN(-1);
+ }
spin_unlock_irqrestore(&imp->imp_lock, flags);
list_for_each_safe(tmp, pos, &imp->imp_sending_list) {
list_for_each_safe(tmp, pos, &imp->imp_delayed_list) {
req = list_entry(tmp, struct ptlrpc_request, rq_list);
- ptlrpc_put_connection(req->rq_connection);
- req->rq_connection =
- ptlrpc_connection_addref(req->rq_import->imp_connection);
-
- if (req->rq_set) {
- DEBUG_REQ(D_HA, req, "waking (set %p):", req->rq_set);
- wake_up(&req->rq_set->set_waitq);
- } else {
- DEBUG_REQ(D_HA, req, "waking:");
- wake_up(&req->rq_wait_for_rep);
- }
+ DEBUG_REQ(D_HA, req, "waking (set %p):", req->rq_set);
+ ptlrpc_wake_client_req(req);
}
spin_unlock_irqrestore(&imp->imp_lock, flags);
}
-inline void ptlrpc_invalidate_import_state(struct obd_import *imp)
+void ptlrpc_request_handle_notconn(struct ptlrpc_request *failed_req)
{
- struct ldlm_namespace *ns = imp->imp_obd->obd_namespace;
- if (ptlrpc_ldlm_namespace_cleanup == NULL)
- CERROR("ptlrpc/ldlm hook is NULL! Please tell phil\n");
- else
- ptlrpc_ldlm_namespace_cleanup(ns, 1 /* no network ops */);
- ptlrpc_abort_inflight(imp);
-}
-
-int ptlrpc_request_handle_eviction(struct ptlrpc_request *failed_req)
-{
- int rc = 0, in_recovery = 0;
+ int rc;
struct obd_import *imp= failed_req->rq_import;
unsigned long flags;
- struct ptlrpc_request *req;
-
- spin_lock_irqsave(&imp->imp_lock, flags);
-
- if (imp->imp_level == LUSTRE_CONN_NOTCONN)
- in_recovery = 1;
-
- if (failed_req->rq_import_generation == imp->imp_generation)
- imp->imp_level = LUSTRE_CONN_NOTCONN;
- else
- in_recovery = 1;
-
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- if (in_recovery) {
- ptlrpc_resend_req(failed_req);
- RETURN(rc);
- }
+ ENTRY;
- CDEBUG(D_HA, "import %s of %s@%s evicted: reconnecting\n",
+ CDEBUG(D_HA, "import %s of %s@%s abruptly disconnected: reconnecting\n",
imp->imp_obd->obd_name,
imp->imp_target_uuid.uuid,
imp->imp_connection->c_remote_uuid.uuid);
- rc = ptlrpc_reconnect_import(imp, &req);
- if (rc) {
- ptlrpc_resend_req(failed_req);
- ptlrpc_fail_import(imp, imp->imp_generation);
- } else {
- spin_lock_irqsave (&failed_req->rq_lock, flags);
- failed_req->rq_err = 1;
- spin_unlock_irqrestore (&failed_req->rq_lock, flags);
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_level = LUSTRE_CONN_FULL;
- imp->imp_invalid = 0;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
- ptlrpc_invalidate_import_state(imp/*, req->rq_import_generation*/);
+
+ if (ptlrpc_set_import_discon(imp)) {
+ if (!imp->imp_replayable) {
+ CDEBUG(D_HA, "import %s@%s for %s not replayable, "
+ "auto-deactivating\n",
+ imp->imp_target_uuid.uuid,
+ imp->imp_connection->c_remote_uuid.uuid,
+ imp->imp_obd->obd_name);
+ ptlrpc_deactivate_import(imp);
+ }
+
+ /* to control recovery via lctl {disable|enable}_recovery */
+ if (imp->imp_deactive == 0)
+ rc = ptlrpc_connect_import(imp, NULL);
}
- ptlrpc_req_finished(req);
- RETURN(rc);
+
+ /* Wait for recovery to complete and resend. If evicted, then
+ this request will be errored out later.*/
+ spin_lock_irqsave(&failed_req->rq_lock, flags);
+ failed_req->rq_resend = 1;
+ spin_unlock_irqrestore(&failed_req->rq_lock, flags);
+
+ EXIT;
}
+/*
+ * This should only be called by the ioctl interface, currently
+ * with the lctl deactivate and activate commands.
+ */
int ptlrpc_set_import_active(struct obd_import *imp, int active)
{
- struct obd_device *notify_obd;
- unsigned long flags;
- int rc;
-
- LASSERT(imp->imp_obd);
+ struct obd_device *obd = imp->imp_obd;
+ int rc = 0;
- notify_obd = imp->imp_obd->u.cli.cl_containing_lov;
+ LASSERT(obd);
- /* When deactivating, mark import invalid, and
- abort in-flight requests. */
+ /* When deactivating, mark import invalid, and abort in-flight
+ * requests. */
if (!active) {
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_invalid = 1;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- ptlrpc_abort_inflight(imp);
- }
-
- imp->imp_invalid = !active;
-
- if (notify_obd == NULL)
- GOTO(out, rc = 0);
-
- /* How gross is _this_? */
- if (!list_empty(¬ify_obd->obd_exports)) {
- struct lustre_handle fakeconn;
- struct obd_ioctl_data ioc_data = { 0 };
- struct obd_export *exp =
- list_entry(notify_obd->obd_exports.next,
- struct obd_export, exp_obd_chain);
-
- fakeconn.cookie = exp->exp_handle.h_cookie;
- ioc_data.ioc_inlbuf1 = (char *)&imp->imp_target_uuid;
- ioc_data.ioc_offset = active;
- rc = obd_iocontrol(IOC_LOV_SET_OSC_ACTIVE, &fakeconn,
- sizeof ioc_data, &ioc_data, NULL);
- if (rc)
- CERROR("error %sabling %s on LOV %p/%s: %d\n",
- active ? "en" : "dis",
- imp->imp_target_uuid.uuid, notify_obd,
- notify_obd->obd_uuid.uuid, rc);
- } else {
- CDEBUG(D_HA, "No exports for obd %p/%s, can't notify about "
- "%p\n", notify_obd, notify_obd->obd_uuid.uuid,
- imp->imp_obd->obd_uuid.uuid);
- rc = -ENOENT;
+ ptlrpc_invalidate_import(imp, 0);
+ imp->imp_deactive = 1;
}
-out:
- /* When activating, mark import valid */
+ /* When activating, mark import valid, and attempt recovery */
if (active) {
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_invalid = 0;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
+ imp->imp_deactive = 0;
+ CDEBUG(D_HA, "setting import %s VALID\n",
+ imp->imp_target_uuid.uuid);
+ rc = ptlrpc_recover_import(imp, NULL);
}
RETURN(rc);
}
-void ptlrpc_fail_import(struct obd_import *imp, int generation)
+int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid)
{
- unsigned long flags;
- int in_recovery = 0;
+ int rc;
ENTRY;
- LASSERT (!imp->imp_dlm_fake);
-
- /* If we were already in recovery, or if the import's connection to its
- * service is newer than the failing operation's original attempt, then
- * we don't want to recover again. */
- spin_lock_irqsave(&imp->imp_lock, flags);
+ /* force import to be disconnected. */
+ ptlrpc_set_import_discon(imp);
- if (imp->imp_level == LUSTRE_CONN_RECOVD)
- in_recovery = 1;
+ rc = ptlrpc_recover_import_no_retry(imp, new_uuid);
- if (generation == imp->imp_generation) {
- imp->imp_level = LUSTRE_CONN_RECOVD;
- imp->imp_generation++;
- } else {
- in_recovery = 1;
- }
+ RETURN(rc);
+}
+int ptlrpc_import_in_recovery(struct obd_import *imp)
+{
+ unsigned long flags;
+ int in_recovery = 1;
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ if (imp->imp_state == LUSTRE_IMP_FULL ||
+ imp->imp_state == LUSTRE_IMP_CLOSED ||
+ imp->imp_state == LUSTRE_IMP_DISCON)
+ in_recovery = 0;
spin_unlock_irqrestore(&imp->imp_lock, flags);
-
- if (in_recovery) {
- EXIT;
- return;
- }
-
- if (!imp->imp_replayable) {
- CDEBUG(D_HA,
- "import %s@%s for %s not replayable, deactivating\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid,
- imp->imp_obd->obd_name);
- ptlrpc_set_import_active(imp, 0);
- }
-
- ptlrpc_run_failed_import_upcall(imp);
- EXIT;
+ return in_recovery;
}
-static int signal_completed_replay(struct obd_import *imp)
+int ptlrpc_import_control_recovery(struct obd_import *imp, int disable)
{
- struct ptlrpc_request *req;
- int rc;
- ENTRY;
-
- req = ptlrpc_prep_req(imp, OBD_PING, 0, NULL, NULL);
- if (!req)
- RETURN(-ENOMEM);
-
- req->rq_replen = lustre_msg_size(0, NULL);
- req->rq_level = LUSTRE_CONN_RECOVD;
- req->rq_reqmsg->flags |= MSG_LAST_REPLAY;
-
- rc = ptlrpc_queue_wait(req);
+ unsigned long flags;
- ptlrpc_req_finished(req);
- RETURN(rc);
+ /* with imp_deactivate == 1 pinger won't initiate re-connect */
+ spin_lock_irqsave(&imp->imp_lock, flags);
+ if (disable)
+ imp->imp_deactive = 1;
+ else
+ imp->imp_deactive = 0;
+ if (imp->imp_state == LUSTRE_IMP_DISCON) {
+ imp->imp_force_verify = 1;
+ ptlrpc_pinger_wake_up();
+ }
+ spin_unlock_irqrestore(&imp->imp_lock, flags);
+ RETURN(0);
}
-int ptlrpc_recover_import(struct obd_import *imp, char *new_uuid)
+static int ptlrpc_recover_import_no_retry(struct obd_import *imp,
+ char *new_uuid)
{
- int msg_flags = 0, rc;
+ int rc;
unsigned long flags;
- struct ptlrpc_request *req;
+ int in_recovery = 0;
+ struct l_wait_info lwi;
ENTRY;
spin_lock_irqsave(&imp->imp_lock, flags);
- if (imp->imp_level == LUSTRE_CONN_FULL) {
- imp->imp_level = LUSTRE_CONN_RECOVD;
- imp->imp_generation++;
+ if (imp->imp_state != LUSTRE_IMP_DISCON) {
+ in_recovery = 1;
}
spin_unlock_irqrestore(&imp->imp_lock, flags);
- if (new_uuid) {
- struct ptlrpc_connection *conn;
- struct obd_uuid uuid;
- struct ptlrpc_peer peer;
- struct obd_export *dlmexp;
-
- obd_str2uuid(&uuid, new_uuid);
- if (ptlrpc_uuid_to_peer(&uuid, &peer)) {
- CERROR("no connection found for UUID %s\n", new_uuid);
- RETURN(-EINVAL);
- }
-
- conn = ptlrpc_get_connection(&peer, &uuid);
- if (!conn)
- RETURN(-ENOMEM);
-
- CDEBUG(D_HA, "switching import %s/%s from %s to %s\n",
- imp->imp_target_uuid.uuid, imp->imp_obd->obd_name,
- imp->imp_connection->c_remote_uuid.uuid,
- conn->c_remote_uuid.uuid);
-
- /* Switch the import's connection and the DLM export's
- * connection (which are almost certainly the same, but we
- * keep distinct refs just to make things clearer. I think. */
- if (imp->imp_connection)
- ptlrpc_put_connection(imp->imp_connection);
- /* We hand off the ref from ptlrpc_get_connection. */
- imp->imp_connection = conn;
-
- dlmexp = class_conn2export(&imp->imp_dlm_handle);
- if (dlmexp->exp_connection)
- ptlrpc_put_connection(dlmexp->exp_connection);
- dlmexp->exp_connection = ptlrpc_connection_addref(conn);
- class_export_put(dlmexp);
+ if (in_recovery == 1)
+ RETURN(-EALREADY);
- }
-
- rc = ptlrpc_reconnect_import(imp, &req);
-
- if (rc) {
- CERROR("failed to reconnect to %s@%s: %d\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid, rc);
+ rc = ptlrpc_connect_import(imp, new_uuid);
+ if (rc)
RETURN(rc);
- }
- if (req->rq_repmsg)
- msg_flags = lustre_msg_get_op_flags(req->rq_repmsg);
+ CDEBUG(D_HA, "%s: recovery started, waiting\n",
+ imp->imp_target_uuid.uuid);
- if (msg_flags & MSG_CONNECT_RECOVERING) {
- CDEBUG(D_HA, "replay requested by %s\n",
- imp->imp_target_uuid.uuid);
- rc = ptlrpc_replay(imp);
- if (rc)
- GOTO(out, rc);
-
- if (ptlrpc_ldlm_replay_locks == NULL)
- CERROR("ptlrpc/ldlm hook is NULL! Please tell phil\n");
- else
- rc = ptlrpc_ldlm_replay_locks(imp);
- if (rc)
- GOTO(out, rc);
-
- rc = signal_completed_replay(imp);
- if (rc)
- GOTO(out, rc);
- } else if (msg_flags & MSG_CONNECT_RECONNECT) {
- CDEBUG(D_HA, "reconnected to %s@%s\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
- } else {
- CDEBUG(D_HA, "evicted from %s@%s; invalidating\n",
- imp->imp_target_uuid.uuid,
- imp->imp_connection->c_remote_uuid.uuid);
- ptlrpc_invalidate_import_state(imp);
- }
-
- rc = ptlrpc_resend(imp);
-
- spin_lock_irqsave(&imp->imp_lock, flags);
- imp->imp_level = LUSTRE_CONN_FULL;
- imp->imp_invalid = 0;
- spin_unlock_irqrestore(&imp->imp_lock, flags);
+ lwi = LWI_TIMEOUT(MAX(obd_timeout * HZ, 1), NULL, NULL);
+ rc = l_wait_event(imp->imp_recovery_waitq,
+ !ptlrpc_import_in_recovery(imp), &lwi);
+ CDEBUG(D_HA, "%s: recovery finished\n",
+ imp->imp_target_uuid.uuid);
- ptlrpc_wake_delayed(imp);
- EXIT;
- out:
- ptlrpc_req_finished(req);
- return rc;
+ RETURN(rc);
}
void ptlrpc_fail_export(struct obd_export *exp)
{
int rc, already_failed;
- struct lustre_handle hdl;
unsigned long flags;
spin_lock_irqsave(&exp->exp_lock, flags);
CDEBUG(D_HA, "disconnecting export %p/%s\n",
exp, exp->exp_client_uuid.uuid);
- hdl.cookie = exp->exp_handle.h_cookie;
- rc = obd_disconnect(&hdl, 0);
+
+ if (obd_dump_on_timeout)
+ portals_debug_dumplog();
+
+ /* Most callers into obd_disconnect are removing their own reference
+ * (request, for example) in addition to the one from the hash table.
+ * We don't have such a reference here, so make one. */
+ class_export_get(exp);
+ rc = obd_disconnect(exp, 0);
if (rc)
CERROR("disconnecting export %p failed: %d\n", exp, rc);
}